* taps the sign *

MIZUHO TO MAKE ANNOUNCEMENT THAT SOUNDS BETTER THAN "WE OVERHIRED WHEN INTEREST RATES WERE LOW"

If you want to know how reliable "AI" coding assistants like Claude Code *really* are before you invest considerable time and $ on one, I highly recommend looking through the Issues on their GitHub repos.

The irony's not lost on me that LLMs are most valuable in the hands of people who don't need them.

When I was in Australia, 2 things happened. It was sunny, and I went for a walk on a farm that was covered in nearly 1 million solar panels. And 6,500 very happy sheep. www.youtube.com/watch?v=TUi0...

Airports that don't spoil my trip

“AI” is an amplifier of engineering capability, not a magic wand that fixes bottlenecks, blockers and quality leaks. Unless you address them (and that .. doesn’t come with your Claude Code plan), it just makes them worse. Thanks @jasongorman.bsky.social 👏 codemanship.wordpress.com/2026/02/26/7...

Met with a friend who works as a developer for a large retailer. I had wine, he had low-alcohol beer. He lives in the Cotswalds now. I found the pub years ago walking through Kennington. If you told somone this before AI, they wouldn't have believed it was possible. #TotalGameChanger

Making America Empty Again

If you want an MP who's actually going to be there, doing the job full-time representing you (and not billionaires), Hannah's the obvious choice.

Dear @lucympowell.bsky.social, Why did you invent a fake tactical voting company to stop the Green Party beating Reform? I'm starting to think your letters haven't always been in good faith...

71% of developers and leaders agree that engineering discipline is more important in AI-assisted and agentic workflows. So why aren't we seeing an increase in investment in engineering capability yet? It doesn't come with your Claude Code plan. codemanship.wordpress.com/2026/02/26/7...

I'd get the money up-front if I were him

Lower bills. Protect the NHS. Rebuild public services. Vote Hannah Spencer & the Green Party.

Labour know they're in third place. Yet in the last 48 hours have gone as far as obsessively attacking the Green Party and even creating a fake tactical voting website to spread misinformation. If you can't vote today - then join us instead! Join.greenparty.org.uk

Move along now; there’s nothing to see here

They’re melting down. We’re building the future. Lower bills. Protect the NHS. Rebuild our public services. Nearly 200,000 members. Join us: greenparty.org.uk

There's no doubting that automating tests is the key to minimising the regression-testing bottleneck. But just because all our tests are green, that doesn't necessarily mean our code's working.

I reckon it could get up to 19 degrees today here. In February.

It's becoming ever clearer that, far from leading to "super-intelligence", scaling LLMs is more likely to end at "super-mediocrity" - producing outputs that are closer to the average of their training data. For software developers, this has implications. codemanship.wordpress.com/2026/02/25/s...

Aww. Sad, pensive cat.

A pincer movement will finish "agentic" coding: * Producing code faster than we can understand it paints us into a corner * The fastest way to understand code is to write it LLM-based tools will ultimately finish up as rubber ducks and inline complete Useful? Sure. $multi-trillion useful? LOL.

So, just to be clear, don't point me at a massively complicated function in a massive change set as your proof that Claude Code produces better code than a good software engineer. I might not stop laughing.

Because the whole point of education is to hand in assignments

The quest for fully autonomous "agentic" software development's increasingly starting to look like alchemy.

Is it cars? I bet it’s cars.

Should software developers care about software architecture? If your answer is "no," we are back in the bad old days of the "Ivory Tower Architect" handing down rigid edicts to developers who are told exactly what to do. 1/9

The correlation is so strong, we could probably measure micromanagement in defects/KLOC

Because they have essentially the same causes, the buggier a code base is, the longer it usually takes to change (because changes keep breaking it). The most popular "AI" coding CLIs seem to be *very* buggy - especially Claude Code. Thousands of open issues. It's also probably very insecure.

That sound you hear is every other intelligence service in the democratic world backing away

Good advice from Kent Beck's TDD book. Dropping into a "lower gear" when your workflow hits a steep gradient's a good option in "AI"-assisted TDD, too. If a problem's outside the model's training data distribution, break it down into smaller steps that are more likely to be in-distribution.

"AI" coding assistants amplify delays and instability caused by more code - with more problems - hitting the bottlenecks in your process at accelerated speed. Will 2026 finally be the year that you address those bottlenecks? codemanship.wordpress.com/2026/02/23/w...

This by-election will be a test of whether the Green Party's messaging can cut through a very hostile mainstream/corporate media

Our legendary trainers: @emilybache.com - Approval tests Lisi Hocke - Secure development Gemma Cameron - tbc @claresudbery.bsky.social (um, me) - TDD + LLMs Jon Jagger - TDD + Cyber-dojo Duncan McGregor & Nat Pryce - Extreme refactoring Tim Ottinger - Value stream mapping @agile-otter.bsky.social

Pssst. Sometimes AI-assistance *is* the hard way. As you were.

Chrome helpfully has a Plausible-Looking Incorrect Answer mode, for when I need to be wrong with authority.

"AI" coding assistants amplify delays and instability caused by more code - with more problems - hitting the bottlenecks in your process at accelerated speed. Will 2026 finally be the year that you address those bottlenecks? codemanship.wordpress.com/2026/02/23/w...

Blade Runner is still thought-provoking and exquisitely beautiful and easily my favourite film. Ridley Scott duped Hollywood into making a $30 million art film. Fun fact: director of photography Jordan Cronenweth was nominated for an Oscar... for Peggy Sue Got Married. Not for Blade Runner.

You often hear "software development is changing, and we must adapt" because of Claude Code. But for my entire career, code has been the *easiest* part of making shit people care about, and we still haven't adapted. The easier coding becomes, the less often we ask "do users actually NEED this?"

This is immoral. Trump officials gutted USAID and cut off food aid to refugees. Clinics closed. Warnings poured in and were ignored. Rations collapsed. Children starved to death. 👉🏾Agency staffers even celebrated the cuts with a sheet cake while lifesaving programs were being shut down.👈🏾

Watching Bringing Up Baby. The speed of the dialogue is incredible.

For some reason, I'm being reminded of when they issued amphetamines to troops in WWII...

"This is why our leaders should be educated in the humanities instead of maths and science", say people who appear to be living in a different reality to me.

Planes are much faster than trains. For me, to get to central Edinburgh by train takes about 5 hours. By plane, about 5 hours. It's not raw speed that determines how long these things take. It's the constraints on the system.

Someone on LinkedIn helpfully educating me on how complexity isn't linked to "clean code"

I'm seeing 2026 turning into a bad year for teams who've been commanded to deliver the "AI" productivity gains their boss read about. When the 10x, 5x, or even 1.5x fails to materialise, teams are getting the blame: "You must be holding it wrong". Heads-up bosses: the "AI" people are lying to you.

Because software engineering is where GenAI is currently having the most visible impact, the AI commentary class have converged on it and are now suddenly experts in SWE productivity, despite little to no experience of actually building and maintaining software.

I'm a big believer in observability in business processes with a significant s/w component. The sooner we know something's amiss, the smaller the impact. But what about observability in software development processes? e.g. How soon would you know that someone on the team isn't running the tests?

I keep hearing that the Epstein files are so big if we held everyone accountable everything would collapse. And no it wouldn’t. These rich guys don’t do any actual work. They aren’t actually important or useful they are just rich.

"Now that code is free..." is our industry's modern equivalent of Ancient Aliens' "Now that we've established it's definitely aliens..."